Built for speed in the driveway
Field-service software lives or dies on how fast it loads while you're standing next to a customer with one bar of LTE. Here are the numbers, and the engineering choices behind them.
~8s
Median AI quote time
From the moment a customer or operator hits Auto-quote to a full itemized estimate. Streamed via Server-Sent Events so progress is visible the whole way.
<1s
Mobile TTI on 3G
The tech day view at /me/today loads and is interactive within one second on a throttled 3G connection, measured in WebPageTest and Lighthouse mobile profiles.
95+
Lighthouse score
Performance, Accessibility, Best Practices, and SEO all in the green on every marketing page and authenticated entry point.
Offline-first
Field workflows
The tech day view ships as a PWA with an IDB-backed outbox. Techs in dead zones queue mutations locally; the service worker drains them on reconnect.
Numbers measured against a representative tenant workload. Your results depend on photo size, network conditions, and the complexity of the property being quoted.
How we keep it fast
Six engineering choices that compound. None of them are magic, they're the boring discipline that adds up to feeling instant.
Streaming where it matters
The AI quote pipeline runs as an async generator: each of the six stages streams progress as it runs. Operators never stare at a spinner wondering if it's working.
Server Components by default
Next.js App Router renders data-heavy pages on the server so the wire payload is HTML, not 2 MB of JavaScript. Interactive bits are isolated client components, not full-page hydration.
Offline-first PWA for the field
Photos, signatures, status flips, payments, and clock-in/clock-out all enqueue locally when offline. The same idempotency key protects against double-replays when the network returns.
Background work, not in the request path
Reminders, ad sync, AI agents, and email/SMS sends run as durable Inngest workflows. Every user request stays snappy because nothing slow happens inside it.
Designed for scale
Tenant isolation runs in Postgres via RLS, queries are bounded by per-tenant limits, and the architecture is ready to shard onto Neon when the volume calls for it.
Measured, not asserted
Performance budgets are tracked, not guessed. Server errors stream to Sentry, product analytics to PostHog, and slow-query monitoring surfaces regressions before customers notice them.
See for yourself
Try the AI quote flow with one of your own property photos and time it. We think you'll be surprised.
Start your free trial